Co-Design Bench

Real-World Test: AI Coding Agents Need Cleanup on ~3 of 4 Tasks

Alibaba's new Co-Design Bench tests coding agents on real user requests that had already tripped them up. Top performer GPT-6 Astra fully solves only 26.4% of tasks, while the same models score above 99% on older benchmarks, highlighting a large gap between benchmarks and real-world use.